Addendum of Digital Video Watermarking MMSEC 2008

Applications of Watermarking Video Content ApplicationsPurpose of the embedded purpose Copy controlPrevent unauthorized copying Broad casting monitoring Identity the video item being broadcasted FingerprintingTrace back a malicious user Video AuthenticationEnsure that the video hasn’t been altered Copyright ProtectionProve IPR ownership Enhanced Video Coding Bring additional information e.g. for error correction

A. Copy Control: DVD( digital versatile disk): late CSS( Content Scrambling System): Matsuhita, MPEG-2 Video A pair of keys is required for descrambling: one is unique to the disk and the other is specific to the MPEG file being descrambled. Scrambled Video is not viewable. CSS was cracked in 1999.

APS( Analog Protection System): Macrovision A modified version of NTSC/PAL video signals. The resulting video signals can be displayed on TV but cannot be recorded on VCR’s. Some bits are stored in the MPEG stream header to give the information of whether and how APS should be applied.

CGMS( Copy Generation Management System) This is a pair of bits stored in the header of an MPEG stream encoding one of the three possible rules for copying: copy-always, copy- never and copy-once. The copy-once case is included so that time- shifting recording is allowed, i.e., a copy of broadcast media is made for later viewing.

5C: A coalition of five companies designs this mechanism. It allows several compliant devices, connected to the same digital video bus, e.g. IEEE 1394 (firewire), to exchange keys in an authenticated manner so that encrypted data can be sent over the bus. Non-compliant devices do not have access to the keys and cannot decrypt the data.

Watermarking: The main purpose of watermarking is to provide a more secure solution than storing bits in the MPEG stream header. In DVD, digital watermarking is primarily intended for the CGMS bits and secondary for the APS bits.

Physical Identifiers: The idea is to design secure physical media identifiers in order to be able to distinguish between original media and copies.

Everything starts when Hollywood studios release a new copyrighted DVD with CGMS bits encoding the message copy-never. Both CSS keys are stored on the “Lean-in” area of the DVD. This area is only read by “ Compliant Players”. This prevents factory- pressed “legal disk” from being displayed by noncompliant players.

Bit-by-bit illegal copies will contain CSS Scrambled content but not the keys. As result, such illegal copies cannot be displayed by any player, compliant or not. If the output signal given by compliant players is digital, CGMS bits prevent copying in the compliant world while 5C will avoid any communication with any non compliant devices.

Nowadays, analog monitors are still widespread and even compliant players output an analog signal for compatibility. Since CGMS bits do not survive digital to analog conversion, watermarking is introduced in order to avoid copying in the compliant world. In the noncompliant world, APS only disables copying of analog NTSC/PAL signals on VHS tapes. Disks without CSS or CGMS can then easily generated ( by using a PC with a video capture card).

Now, illegal disks containing unscrambled content without CSS and CGMS are available. They may have been generated as prescribed or they may also be generated directly from an original legal disk since CSS was cracked in The remaining CGMS bits can then be trivially stripped from the MPEG stream. DeCSS gives clear MPEG streams.

Such illegal copies can of course be displayed by noncompliant players but watermarking has to be introduced in order to prevent those copies to enter the compliant world. Compliant players will detect the copy-never watermark embedded in unscrambled DVD- ROM and refuse playback.

The video signal given by a noncompliant player can be recorded by noncompliant recording devices. However, watermarking prevents copying with compliant devices. The whole protection system results in two hermetically separated worlds. A consumer should have both types of players in order to display legal and illegal disks. The expense of such a strategy will help to “ keep honest people honest”.

Legal Disk (CSS) Illegal Disk (CSS) Illegal Disk (No CSS) Compliant Player Non-Compliant Player CSS Playback Control CSS APS,5C CGMS Recordin g control Compliant recorder Non-compliant record DVD Copy-protection System

B. Broadcasting monitoring Passive monitoring: Human or computer simulated observer monitors the broadcasts and compares the received signals with a database of known videos. Active monitoring: Transmit computer recognizable identification information(i.e. the watermark) along with the broadcasting videos.

C. Fingerprinting: Download video content over Internet through peer-to-peer systems By embedding an indelible and invisible watermark to identify the traitor when an illegal copy is found. The technique is getting more attention recently because of the popularity and maturity of video streaming applications, such as PPV (Pay-per- View) and VOD(Video-on-Demand).

In both applications, digital watermarking (user ID) is embedded into the delivered video data in order to trace back any user breaking his/her license agreement. Embedding the watermark on the customer side has been suggested but it should be avoided if possible in order to prevent reverse engineering.

In a ppv environment, a video server multicasts some videos and customers have only to connect to the server in order to obtain the video. The video server is passive. In order to enforce fingerprinting, one has to embed a piece of watermark to each network element (router, node or whatever ) as the video stream is related. The resulting watermark will contain a trace of the route followed by the video stream. Such a strategy requires the support from network providers.

In a VDD framework, the video server is active. It receives a request from a customer and sends the required video. It is a multi-unicast strategy. This time, the video server can insert a watermark identifying the customer since each connection is dedicated to only one customer. The main challenge is then to scale the system to many users.

Nowadays, illegal copying of brand new movies projected onto cinema screen by means of a handheld video camera has become a common practice ---- with visible head shadows of audience members. A watermark can be embedded during show time identifying the cinema, the presentation date and time. If an illegal copy created with a video camera is found, the watermark is extracted and the cinema to blame is identified. After many blames, the cinema is sanctioned with a ban on the availability of content.

Popular Video Editing softwares permit to easily temper with video content so that video content is no more reliable. In some countries, a video shot taken from a surveillance camera cannot be used as a piece of evidence in a courtroom because it is not considered trust- worthy emough. Authentication techniques are needed to ensure authenticity of video content. D. Video Authentication

Video authentication might prefer to allow some distortions on the digital data if the original content has not been significantly modified: fragile vs. Semi- fragile watermarks! A basic approach consists in regularly embedding an incremental timestamp to the frames of the video. As a result, frame cuts, foreign frame insertion, frame swapping, and frame rate alteration can be easily detected temporal alternation of the video stream.

The above approach might fail in detecting alternations of the content itself, e.g., a character is completely removed from a movie. One proposal embeds the edge map of each frame to the video stream. During the verification process, if the video content has been modified, there will be a mismatch between the extracted edge map from the verified video and the watermarked edge map.

Another proposal exploits the fact that a movie is made up of one audio and one video stream and that both need to protected against unauthorized tempering. By combining video and audio watermarking to obtain an efficient authentication system. Modifications from either the sound track, or the video track, is immediately spotted by the detector, since the extracted and watermarked features will differ.

historically the very first target applications for digital watermarking. Deadlock problem: If an attacker adds a second watermark into a video clip, both the original owner and the attacker can claim ownership and therefore defeat the purpose of using watermark a trusted third party is required to build an application watermarking protocol.( Buyer-Seller Protocol.) E. Copyright Protection :

Video watermarking and video coding are two conflicting technologies. However, recent research has shown that digital watermarking can benefit to the coding community. The video coding process can be sequenced in two steps: Source coding : any redundant information is removed in order to obtain the most possible compressed representation of the dada while keeping its original visual quality. F. Enhanced Video Coding

Channel coding: extra redundant information is added for error correction. Digital watermarking can be introduced as an alternative solution for introducing error correcting information after source coding, without inducing any overhead. Experiments have demonstrated that digital watermarking can have a better error correcting performance than traditional error correction mechanisms.

Embedding useful data directly into the video stream can spare much storage space. A typical video stream is made up of two different parallel streams : the audio and video streams. Those two streams need to be synchronized during playback for pleasant viewing, which is difficult to maintain during cropping operations. Hiding the audio stream into the video one will implicitly provide efficient and robust synchronization, while significantly reducing the required storage need or available bandwidth.

Picture-in-picture system can be improved by hiding a video stream into another one. During playback, the watermark is extracted and the embedded video is displayed in a window within the host video. With such an approach, only one stream needs to be transmitted. This approach can be extended so that a user can switch to the PG version of an R-rated movie, with alternative dialogs and senses replacing inappropriate content.

There are 3 major challenges for digital video watermarking. (1) there are many non-hostile video processing, which are likely to alter the watermark signal. (2) resilience to collusion is much more critical in the content of video. (3) real-time digital video watermarking [II]. Challenges in Video Watermarking

(i) Photometric attack : the attacks modify the pixel values in the frame. ‧ Noise addition, DA/AD conversion ‧ Gamma correction ‧ Trans-coding and video format conversion ‧ Intra and Inter-frames filtering ‧ Chrominance resampling (4:4:4, 4:2:2, 4:2:0) a.Various non-hostile Video Processing

(ii) Spatial desynchronization : many watermarking algorithms rely on an implicit spatial synchronization between the embedder and the detector. ‧ Changes across display formats (4/3, 16/9, 2.11/1) ‧ Changes of spatial resolution (NTSC, PAL, SECAM) ‧ Positional jitter ‧ Handheld camera attack (video stabilization) (iii) Temporal desynchronization : ‧ Changes of frame rate

(iv) Video editing : ‧ Cut-and-splice and cut-insert-splice ‧ Fade-and-dissolve and wipe-and-matte ‧ Graphic overlay (subtitles, logo) : Commercial insertion

A set of malicious users who merge their knowledge, i.e., different watermarked data, in order to produce illegal content, i.e., unwatermarked data. Collusion Type I: The same watermark is embedded into different copies of different data. The collusion can estimate the watermark from each watermarked data and obtain a refined estimate of the watermark by linear combination, e.g. the average, of the individual estimations. Having a good estimate of the watermark permits to obtain unwatermarked data with a simple subtraction with the watermarked one. b. Resilience against Collusion

Collusion Type II: Different watermarks are embedded into different copies of the same data. The collusion only has to make a linear combination of the different watermarked data, e.g. the average, to produce un watermarked data. Indeed, generally, averaging different watermarks converges toward zero.

Inter-videos Collusion: A set of users have a watermarked version of a video which they gather in order to produce unwatermarked video content. Intra-video Collusion: Watermarking video comes down to watermarking series of still images.

In order to meet the real-time requirement, the complexity of the watermarking algorithm should obviously be as few as possible. Moreover, if the watermark can be inserted directly into the compressed stream, this will prevent full decompression and recompression consequently, it will reduce computational needs. — VLCs based Video Labelling. C. Real-time Watermarking